- Uber AI costs are rising fast, but COO Andrew Macdonald says there’s no clear link to better consumer features.
- Uber already burned through its entire Claude Code budget for 2026 by April, sparking internal debate about AI spending.
- CEO Dara Khosrowshahi confirmed Uber is slowing hiring to offset growing investment in AI tools.
- Duolingo also reversed a policy tying performance reviews to AI usage, signalling a wider industry rethink.
- Uber AI costs are rising fast, but COO Andrew Macdonald says there’s no clear link to better consumer features.
- Uber already burned through its entire Claude Code budget for 2026 by April, sparking internal debate about AI spending.
- CEO Dara Khosrowshahi confirmed Uber is slowing hiring to offset growing investment in AI tools.
- Duolingo also reversed a policy tying performance reviews to AI usage, signalling a wider industry rethink.
Uber AI Costs Are Climbing — and the C-Suite Is Asking Why
Uber AI costs have become a boardroom-level headache, and the company’s own operations chief isn’t hiding it. In a Rapid Response podcast interview published this past weekend, COO Andrew Macdonald said it’s getting genuinely harder to justify the money Uber is pouring into AI — and that the internal conversation around this has turned unexpectedly heated.
The trigger was a candid comment from Uber CTO Praveen Neppalli Naga, who told The Information in April that Uber had already exhausted its entire Claude Code budget for 2026. That’s Anthropic’s AI-powered coding assistant, and burning through a full-year allocation in roughly four months is the kind of detail that tends to cause what Macdonald called a “head-exploding moment” among senior leadership. It did exactly that.
The fallout wasn’t just internal embarrassment. It forced a serious conversation about what Uber is actually getting for all that spend — a question that, surprisingly, doesn’t have a clean answer right now.
More Tokens, Not More Features
Here’s where it gets interesting. After sitting down with Uber’s senior engineering leaders, Macdonald came away unconvinced that higher AI token consumption is translating into meaningfully better products. The correlation simply isn’t there yet.
“That link is not there yet, right?” Macdonald said in the interview. “I think maybe implicitly there is more that is getting shipped, but it’s very hard to draw a line between one of those stats and, ‘Okay, now we’re actually producing 25% more useful consumer features.'”
That’s a remarkably honest admission from a senior executive at a company that, like most of Silicon Valley, has been under enormous pressure to show it’s all-in on AI. The expectation — from investors, from the press, from the broader tech culture — is that AI adoption equals productivity gains equals competitive advantage. Macdonald is essentially saying: we’re not sure that equation holds, at least not yet, at least not for us.
Uber AI costs aren’t just a line item. They come with trade-offs. Every dollar directed toward AI infrastructure and tooling is a dollar not going toward headcount, infrastructure, marketing, or driver incentives. When those trade-offs are easy to justify — because the productivity gains are obvious and measurable — nobody complains. When they’re not, the whole logic starts to unravel.
The Headcount Question
The timing of all this matters. Just earlier this month, Uber CEO Dara Khosrowshahi told investors on an earnings call that the company is deliberately slowing hiring to help offset its AI investments. That’s not a subtle move. Slowing headcount growth at a company of Uber’s scale — roughly 32,000 employees globally — carries real consequences for product velocity, operations, and morale.
The implicit promise behind that trade-off is that AI will fill the gap: that engineers using tools like Claude Code will ship faster, that support teams using AI assistants will handle more volume, that fewer humans can do more. But if Macdonald is right — if Uber AI costs aren’t yet producing a proportional return — then the company may be walking a precarious line between genuine efficiency and self-imposed constraint.
Macdonald made another observation worth sitting with. He pointed out that AI can feel practically free when you’re an individual user experimenting with prompts and ideas. The cost is invisible. But when you’re the company paying the API bills, the infrastructure costs, and the licensing fees, the reality is very different. Scale changes everything. What looks like a free productivity tool at the individual level becomes a significant budget line when multiplied across thousands of engineers running thousands of queries every day. Uber AI costs illustrate exactly how quickly that invisible overhead compounds at enterprise scale.
Is ‘Tokenmaxxing’ the New Vanity Metric?
There’s a broader trend worth naming here. A growing number of tech companies have adopted what some are calling “tokenmaxxing” — essentially, maximising AI usage across the organisation as a strategic goal in itself. Big Tech players like Google, Microsoft, and Meta have leaned hard into this, and some have reportedly begun evaluating employees partly on the basis of how much they’re using AI tools. The logic is straightforward: if you’re not using AI, you’re falling behind.
But Uber AI costs are starting to illustrate the problem with treating token consumption as a proxy for productivity. Usage isn’t output. Activity isn’t value. A team that burns through a massive Claude Code budget but ships the same number of features isn’t winning — it’s just spending more.
Uber isn’t alone in pushing back. Duolingo — which made headlines earlier this year when it announced it was becoming an “AI-first” company and would factor AI usage into employee performance reviews — has since walked that back. CEO Luis von Ahn admitted in an April podcast interview that the policy felt wrong in practice.
“It felt like, rather than being held accountable for the actual outcome, we were trying to just push something that in some cases did not fit,” von Ahn said.
That’s a meaningful retreat. Duolingo built a lot of its public identity around aggressive AI adoption this year, so reversing course on a core policy sends a signal. Employees were essentially asking: do we have to use AI even when it doesn’t help? And the honest answer turned out to be no.
What ROI Actually Looks Like for Enterprise AI
The challenge facing Uber — and many companies like it — is that measuring AI ROI remains genuinely difficult. McKinsey’s ongoing research into enterprise AI adoption consistently finds that organisations struggle to connect AI investments to measurable business outcomes, particularly in areas like software development, where the causal chain between a developer using an AI coding assistant and a customer-facing feature actually shipping is long and indirect.
Uber AI costs are high partly because the company is operating at massive scale across a complex, real-time logistics platform. The stakes for reliability are high, which means engineering teams can’t just ship AI-generated code without thorough review. That review process takes time and human judgment — which somewhat undermines the efficiency argument for AI coding tools in the first place.
None of this means AI isn’t valuable. Uber almost certainly is getting something for its investment — faster prototyping, better documentation, reduced friction on routine coding tasks. But “something” isn’t the same as the transformative productivity leap that the broader industry narrative has been promising. And Macdonald deserves credit for saying so publicly, at a moment when admitting AI uncertainty still carries a certain reputational risk in tech circles. The pressure to show results is precisely why Uber AI costs have become such a flashpoint internally.
A Correction the Industry Probably Needs
What’s playing out at Uber may be the beginning of a broader recalibration. The 2023–2025 period saw an almost religious fervour around AI adoption — spend more, use more, automate everything, and trust that the returns will follow. That faith is starting to collide with finance departments demanding receipts.
Uber AI costs burning through a full-year budget in four months isn’t a catastrophe. But it is a forcing function. It makes leadership ask the right questions: What are we actually getting? Where is this working? Where is it theatre? Those questions should have been asked from the start, but better late than never.
If more companies follow Uber and Duolingo in demanding outcome-based accountability for AI spending — rather than rewarding raw usage — it could push AI vendors to get much better at demonstrating real-world value. Scrutiny over Uber AI costs, in that sense, may end up doing the broader industry a favour. That pressure, in turn, might be exactly what’s needed to move past hype and into something more durable.
Source: https://www.businessinsider.com/uber-coo-andrew-macdonald-ai-token-spending-harder-justify-2026-5
